The Twenty Minute VC (20VC): Venture Capital | Startup Funding | The Pitch

Episode Title: Enterprises Will Not Adopt AI without Forward-Deployed Engineers | Who Wins the Data Labelling Race: How Does it Shake Out? | How Synthetic Data Threatens the Future of Human-Generated Data

Guest: Matt Fitzpatrick, CEO of Invisible Technologies

Host: Harry Stebbings
Date: December 31, 2025

Episode Overview

This episode features a deep-dive conversation between Harry Stebbings and Matt Fitzpatrick, CEO of Invisible Technologies. The discussion explores the chasm between enterprise AI adoption and model performance, the economics and necessity of forward-deployed engineers, the nuances in data labeling and AI training platforms, and the evolving future of data, including the roles of synthetic and human-generated datasets. Fitzpatrick, drawing on his background at McKinsey (QuantumBlack) and his leadership at Invisible, provides candid insights on go-to-market strategies, enterprise pain points, industry structure, and leadership in disruptive times.

Episode Highlights & Key Discussion Points

1. Matt Fitzpatrick’s Background & Journey to Invisible (04:53–07:38)

• McKinsey & QuantumBlack:
  • Spent 12 years at McKinsey, rising to Senior Partner and leading QuantumBlack Labs.
  • Oversaw major scaling from 100 to 7,000 engineers, focused on application development, data warehouse infrastructure, and global gen AI solutions.
• Transition:
  • Built ties with Invisible’s founder, Francis, through personal forums focused on ideas, not work.
  • Recruited to Invisible to scale the company in the U.S. as CEO.
• Decision-Making:
  • “When you walk away from a really stable job that you really enjoy, that's always difficult… But to run a company in AI right now is fascinating.” (06:46)

2. Decision-Making Frameworks (07:38–09:07)

• Personal Philosophy:
  • Driven more by enjoyable work with smart people than by material outcomes.
  • Leans on advice from trusted mentors and loved ones.
  • Memorable advice: “The only risk is if you don’t take this and the amount of regret you’ll have not give it a go.” (09:07)

3. The Enterprise AI Adoption Challenge (09:24–11:19)

• Model Progress vs. Enterprise Stagnation:
  • Consumer GenAI adoption is high (60% weekly, KPMG), while enterprise deployments lag (MIT: 5% of enterprise deployments working).
  • Enterprise hurdles: data infrastructure, workflow redesign, accountability, trust, observability.
• Key Insight:
  • "The deployment of the Enterprise is a lot more than just models themselves... And so I think that whole process is in the first inning in the Enterprise. I think it’s going to take a decade, not two years." (10:42)

4. Enterprise Build vs. Buy: Internal vs. External AI (12:03–15:19)

• Financial discipline and expertise in external vendors far outpaces internal teams in many enterprises.
• “Externally driven builds are 2x as effective as internal team builds.” (12:14)
• Enterprises risk science project dead-ends due to lack of discipline and qualified talent.
• Vivid example: An e-commerce returns agent built for $25M was abandoned after it failed real-world tests.

5. Practical Advice for CEOs & CFOs (15:19–18:06)

• You don’t need a highly technical CFO; you need operational discipline, viable data, outcome-based milestones, and proper incentive alignment.
• Let the business, not just tech, run GenAI initiatives. "Don't locate GenAI in the tech function. ...Your Genai initiative should be led by the business and figure out that could be your head of call center, …with clear operational KPIs." (16:33)

6. Proliferation of Vendors and Enterprise Confusion (18:06–19:20)

• Market flooded with 250+ vendors/week pitching undifferentiated solutions.
• Out-of-the-box agents: just 58% accurate in single turn, 33% in multi-turn workflows—“they don’t really work.”
• Quote: “Start with proof of concepts… don’t pay a dollar until you prove the tech works.” (19:03)

7. Invisible Technologies: Product Architecture & Deployment Model (19:22–21:52)

• Five core modules: Neuron (data), Axon (agent builder), Atomic (process builder), expert marketplace, Synapse (evaluation).
• Use case discussed: Concierge medicine—integrating EHR, CRM, wearables, automating workflows, and building hyper-personalized software rapidly compared to legacy consultancies.

8. Forward-Deployed Engineers (FDEs) as the Key to AI Adoption (21:52–24:56)

• FDEs are non-negotiable for real enterprise transformation—"You just cannot do this with out-of-the-box SaaS. It does not work." (21:52)
• Invisible invests in FDEs over sales, deploying small teams for fast, configurable builds.
• Enterprise startups must consider FDEs for workflow-changing products.

9. The Changing SaaS & Enterprise AI Pricing Model (25:37–27:56)

• Traditional SaaS margins collapse when customization and hands-on implementations are required.
• New paradigm: Payment only after real “user acceptance testing and validation.”
• “We use SaaS as a paradigm because that's how software has worked. But … the GenAI adoption paradigm in the enterprise works the same way that ML did.” (27:56)

10. Data Labeling, Human Expertise, & Marketplace Structure (27:56–33:13)

• Talent Marketplace vs. AI Training Platform:
  • Invisible combines a global expert marketplace with proprietary training/fine-tuning IP.
• Specialization & Niche Data**:
  • Market shifting from "cat-dog" labeling to hyper-specialized tasks (e.g., “validate an expert on 17th-century French architecture who speaks French in 24 hours”).
• Human-generated nuanced data remains invaluable for highly complex tasks, especially enterprise-grade and context-specific reasoning.

11. Commoditization, Market Structure & Competitive Dynamics (33:13–41:48)

• Data is not yet commoditized—quality, complexity, and specialization create defensibility.
• Market will likely maintain 3–5 significant players rather than consolidate to one.
• Companies like Palantir are respected for early insight into need for forward-deployed engineering and deep customization.

12. The Synthetic Data Debate & Human in the Loop (49:33–51:00)

• The recurring narrative that synthetic data will replace human-labeling is overstated.
• For complex, contextual, culturally nuanced tasks, “the only way to actually do the fine-tuning process consistently ... is RLHF [Reinforcement Learning with Human Feedback].” (49:33)
• Enterprises will require humans-in-the-loop for “decades to come.”

13. Growth vs. Profitability, Investment, and Physical World Data (51:00–53:13)

• Invisible is investing heavily in technology (“greatest environment for growth that has ever existed”).
• New frontiers: integrating physical-world data (agriculture, oil/gas, robotics) involves physical deployments and on-site engineering.
• “Brand was historically underplayed; trust and credibility are core to long-term value.”

14. Organization & Culture (61:07–66:54)

• Recruitment focus: Hire all-around athletes for evolving roles and maintain enjoyment to ensure retention.
• Culture: Research/exploration-driven firms require a different approach than hyper-efficient execution cultures. “You have to create an environment where people really enjoy going to work every day, where they're intellectually challenged and where they feel like they can unleash creativity.” (62:07)
• Shift to in-person offices (after years fully remote) has driven stronger culture and serendipity.

15. Management Philosophy & Adaptation (68:36–71:14)

• "Control is a fallacy" in high-growth, high-complexity businesses: Flatten hierarchy and empower teams at the edge.
• In AI, five-year strategic planning is obsolete; adaptability and iteration are paramount.

16. Personal Insights, Work-Life Balance, and Leadership (71:14–73:13)

• Fitzpatrick balances frequent travel with relationships by prioritizing partner understanding and leveraging digital connectivity.
• Most valuable CEO advice: “Recruit great people, create a culture…, build great things, and make them all extremely rich.” (73:18)

Notable Quotes & Memorable Moments

• On Enterprise AI Reality:
  “Enterprise is in the first inning. I think it’s going to take a decade, not two years.” (10:42, Matt Fitzpatrick)

• On Internal Builds:
  “Externally driven builds are 2x as effective as internal team builds.” (12:14, Matt Fitzpatrick)

• On Data Value:
  “People are willing to pay for good data. …If you win and your data is way better, people are willing to pay for that.” (31:24, Matt Fitzpatrick)

• On Forward-Deployed Engineers:
  “For deployed engineers are the only way to do it… It is the only way to get the enterprise working.” (25:00, Matt Fitzpatrick)

• On Agents:
  “The biggest misconception now is that… out-of-the-box agents will solve everything with a push of a button.” (73:53, Matt Fitzpatrick)

• On the Human-In-the-Loop Future:
  “On the Gen AI side you are going to need humans in the loop for decades to come.” (49:33, Matt Fitzpatrick)

• On Management Evolution:
  “Control is a bit of a fallacy depending on the volume of things you have going on… you need to empower all those teams at the edge.” (68:42, Matt Fitzpatrick)

Timestamps for Key Segments

• Matt’s professional background & journey: 04:53–07:38
• Decision-making philosophy & mentors: 07:38–09:07
• Enterprise AI adoption challenges: 09:24–11:19
• Build vs. buy for enterprise AI: 12:03–15:19
• Advice for CEOs & operational leadership: 15:19–18:06
• Vendor fatigue & solution validation: 18:06–19:20
• Invisible’s modular product platform: 19:22–21:52
• The necessity & economics of FDEs: 21:52–24:56
• Enterprise SaaS and pricing evolution: 25:37–27:56
• Expert platform, specialization, & data value: 27:56–33:13
• Commoditization, supply, and market structure: 33:13–41:48
• Synthetic vs. human data for the next decade: 49:33–51:00
• Culture, management, and org structure: 61:07–66:54

Episode Flow & Tone

The conversation is candid, unhurried, and full of practical wisdom. Matt Fitzpatrick combines granular, technical expertise with strategic and philosophical takes on technology change management. Harry Stebbings provides accessible, sometimes irreverent, prompts, keeping the tone lively and relatable even as the topics dig into deep enterprise tech, economics, and product-market strategy.

For Listeners Who Haven’t Tuned In

This episode is a practical masterclass for startup founders, enterprise leaders, and VCs interested in the untold pain points of real-world AI deployments. You’ll hear how to cut through vendor noise, why incremental SaaS paradigms fall short for GenAI, how to structure successful enterprise AI projects, and why human expertise in data and workflow remains a lasting competitive advantage. Fitzpatrick's war stories—like the failed $25M internal agent build, his priorities as a new CEO, and balancing AI optimism with realism—add color and credibility.

You’ll leave with a realistic, actionable understanding of enterprise AI’s current bottlenecks, the real economics and workflows behind successful implementation, and what qualities matter in the next era of technology leadership.